Classifying mental workload using EEG data: A machine learning approach

Harputlu Aksu, Şeniz; Çakıt, Erman · 2022 · Crossref

DOI: 10.54941/ahfe1001820

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the relationship between electroencephalography (EEG) features, task difficulty, and subjective workload ratings, while developing machine learning algorithms to classify mental workload levels. The research was motivated by the need for objective, real-time physiological measures to assess cognitive load in human-machine interactions, as excessive workload leads to processing errors. The authors aimed to address the known difficulty in classifying multiple workload levels by testing both traditional and newer gradient-boosting algorithms. The experimental design involved four participants (mean age 36) performing n-back memory tasks at four difficulty levels (0-back to 3-back) using Millisecond software. EEG data was recorded via an EMOTIV EPOC X device (14 channels, 128 Hz sampling rate) following the international 10-20 system. Each session comprised 12 blocks, with participants completing the NASA-TLX scale after each block to provide subjective workload data. For analysis, 70 EEG features were extracted, consisting of power in five frequency bands (4–8, 8–12, 12–16, 16–25, and 25–45 Hz) across 14 channels. Statistical analysis utilized Spearman correlation and Kruskal-Wallis tests, while classification models included k-Nearest Neighbors (kNN), Support Vector Machines (SVM), Artificial Neural Networks (ANN), Random Forest (RF), Gradient Boosting Machines (GBM), LightGBM, and XGBoost. Results indicated a strong positive correlation between task difficulty and weighted NASA-TLX scores (rho: 0.84). EEG analysis revealed that theta power in prefrontal and frontal regions increased with task difficulty, while high-beta power in these areas decreased. Conversely, gamma power in occipital, parietal, and temporal regions increased with perceived workload. In the four-class classification problem, the Random Forest algorithm achieved the highest accuracy of approximately 68%, with misclassifications primarily occurring between adjacent difficulty levels (0-1 and 2-3). When the problem was simplified to a two-class model distinguishing low versus high workload, the GBM algorithm achieved the maximum accuracy of approximately 87%. Feature importance analysis identified prefrontal/frontal theta, prefrontal high-beta, and occipital/parietal/temporal gamma as the most critical predictors. The significance of this work lies in demonstrating that EEG signals alone are effective for estimating mental workload, particularly when using recent gradient-boosting algorithms like GBM, LightGBM, and XGBoost, which outperformed or matched traditional methods. The findings suggest that while four-level classification remains challenging, binary classification of workload states is highly feasible. The study highlights the utility of these newer algorithms in neuroergonomics research and suggests that increasing sample sizes and refining feature selection could further improve multi-class classification performance.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data